Skip to content
Prev 57842 / 63421 Next

Inconsistencies in wilcox.test

I see I am too late to comment :)

But commenting after the fact, just wish to say that I like the changes. 
Specially the mentioning of "exact" in the test name.

Floating point prevision is very nicely implemented too.
My only worry is that it will not serve new/lay users that may be in the 
biggest need for protections like these.

Do you think it would make sense to do it a bit differently? i.e.
setting digits.rank=7 by default, and including a message in the warning 
i.e. "ties present, if you are working with small digits consider 
adjusting digits.rank".

But, on the other hand, I understand that this would be a breaking 
change. A non breaking change might be to leave digits.rank as NA or 
NULL by default, which would act as infinity but also would do a test 
within wilcox.test() that checks for ties with digits.rank=7. Then a 
warning will say "possibly missed ties due to machine precision, if you 
are sure these are not ties - set digits.rank to Inf to get rid of this 
warning". This would be a non-breaking change, except for a warning. 
Would be interesting to hear your thoughts about this.

I will pull your changes and try to play with the code a bit later 
today. Thanks a lot for, Martin!

Also I have an unrelated question - I mainly find these discrepancies in 
"stats" because I am working on my little package related to hypothesis 
tests. And I have found a few more of them in other tests. One that I 
reported long time ago, regarding flinger.test(), which is also related 
to machine precision.

In terms of the etiquette of this list - should I mention them in this 
same thread or is it better to create a new one?

Kind regards,
Karolis K.
On 2019-12-14 22:50, Martin Maechler wrote: